A Meta-Analysis of Massage Therapy Research

Christopher A. Moyer, James Rounds, and James W. HannumUniversity of Illinois at Urbana–Champaign

Massage therapy (MT) is an ancient form of treatment that is now gaining popularity as part of thecomplementary and alternative medical therapy movement. A meta-analysis was conducted of studiesthat used random assignment to test the effectiveness of MT. Mean effect sizes were calculated from 37studies for 9 dependent variables. Single applications of MT reduced state anxiety, blood pressure, andheart rate but not negative mood, immediate assessment of pain, and cortisol level. Multiple applicationsreduced delayed assessment of pain. Reductions of trait anxiety and depression were MT’s largest effects,with a course of treatment providing benefits similar in magnitude to those of psychotherapy. Nomoderators were statistically significant, though continued testing is needed. The limitations of a medicalmodel of MT are discussed, and it is proposed that new MT theories and research use a psychotherapyperspective.

Massage therapy (MT), the manual manipulation of soft tissueintended to promote health and well-being, has a history extendingback several thousand years. Recorded in writing as far back as2000 B.C. (Fritz, 2000, p. 13), massage was a part of many ancientcultures including that of the Chinese, Egyptians, Greeks, Hindus,Japanese, and Romans, and was often considered to be a medicinalpractice (Elton, Stanley, & Burrows, 1983, p. 275). The Greekphysician Hippocrates (460–377 B.C.) advocated rubbing as atreatment for stiffness; later, the physicians Celsus (25 B.C.–A.D.50) and Galen (A.D. 129–199) wrote extensively on the medicinaland therapeutic value of massage and related techniques such asanointing, bathing, and exercise. However, in Western cultures, theassociation between massage and medicine eventually diminishedas Greco-Roman traditions were abandoned. Although the practiceof massage continued as a folk medicine treatment during theMiddle Ages, its adoption by the common people served to sepa-rate it from the scientific and medical milieu, and in this way,massage fell out of favor with the medical establishment (Fritz,2000; Salvo, 1999).

This schism continued during the early part of the 19th century,during which time Per Henrik Ling developed Swedish massage,the basis of many modern forms of MT. Ling, who was not trainedin medicine, applied his ideas and techniques to the treatment ofdisease, a practice that met opposition from the Swedish medicalcommunity. Despite this resistance, Ling gained support from hisinfluential clients and was eventually able to teach his system to

physicians, who adopted his techniques and shared them withlike-minded colleagues. Soon after, in the later part of the century,the Dutch physician Johann Mezger was successful in reintroduc-ing massage to the scientific community, presenting it to hiscolleagues as a medical treatment, and codifying some of itselements with terms that are still in use today (Fritz, 2000, pp.16–17; Salvo, 1999, pp. 9–11).

Interest in MT has continued to grow among the scientificcommunity and consumers alike. Currently, in the United States,MT is one of the fastest growing sectors of the expanding com-plementary and alternative medical therapy movement. Visits tomassage therapists increased 36% between 1990 and 1997, withconsumers now spending between $4 and $6 billion annually forMT (Eisenberg et al., 1998), in pursuit of benefits such as im-proved circulation, relaxation, feelings of well-being, and reduc-tions in anxiety and pain, all of which are endorsed as benefits ofMT by the American Massage Therapy Association (AMTA,1999b). At the same time, numerous studies across several fieldsincluding psychology, medicine, nursing, and kinesiology supportMT’s therapeutic value. Field (1998) reviewed the effectiveness ofMT in treating symptoms associated with a host of clinical con-ditions, including pregnancy, labor, burn treatment, postoperativepain, juvenile rheumatoid arthritis, fibromyalgia, back pain, mi-graine headache, multiple sclerosis, spinal cord injury, autism,attention-deficit/hyperactivity disorder, posttraumatic stress disor-der, eating disorders, chronic fatigue, depression, diabetes, asthma,HIV, and breast cancer. In addition to the beneficial outcomes thatwere unique to these specific conditions, Field proposed a set ofcommon findings by indicating that “across studies, decreaseswere noted in anxiety, depression, [and] stress hormones (corti-sol)” (p. 1278).

Even the popular press has picked up on the increase in MTpractice and research. A feature in Time suggested that MT is onthe rise, in part, because of “people’s greater awareness of theeffect stress has on health” (Luscombe, 2002, p. 49). It is alsoreported that the National Institutes of Health have begun fundingMT research, and that the White House Commission on Comple-mentary and Alternative Medicine Policy (2002) has called for

Christopher A. Moyer, James Rounds, and James W. Hannum, Depart-ment of Educational Psychology, University of Illinois at Urbana–Champaign.

We wish to thank Sue Duval, Carol Webber, and the InterlibraryBorrowing Staff at the Illinois Research and Reference Center, Universityof Illinois at Urbana–Champaign, for their invaluable contributions to thisproject. Patrick Armstrong and James Wardrop also contributed.

Correspondence concerning this article should be addressed to JamesRounds, Department of Educational Psychology, University of Illinois atUrbana–Champaign, 1310 South Sixth Street, Champaign, IL 61820-6990.E-mail: jrounds@uiuc.edu

Psychological Bulletin Copyright 2004 by the American Psychological Association, Inc.2004, Vol. 130, No. 1, 3–18 0033-2909/04/$12.00 DOI: 10.1037/0033-2909.130.1.3

3

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

more research and public education on MT. The Time articleconcludes by noting that the Commission’s chairman, physicianJames Gordon, indicates that MT is known to be effective indecreasing anxiety, reducing pain, and improving mood (Lus-combe, 2002, p. 50).

If MT can be effective in the ways indicated by the AMTA,Field, and Gordon, it would represent a therapy of interest to avariety of fields. One can imagine its use expanding beyond theprivate practices of massage therapists, and extending to placessuch as hospitals, nursing homes, psychological treatment centers,sports performance clinics, and workplaces. In addition, MT couldestablish itself as a treatment supported by insurance carriers andhealth maintenance organizations. These are, in fact, trends that arealready occurring in a limited way. Nevertheless, for these trendsto continue (indeed, to determine if they even should continue),what is needed is a more rigorous and quantitative examination ofMT’s effectiveness than that which currently exists.

There are three meta-analyses of MT research, but each is verylimited in scope. Ottenbacher et al. (1987) quantified 19 studiesthat examined the effects of tactile stimulation on infants andyoung children, and found statistically significant beneficial out-comes for five of the six categories examined: motor–reflex,cognitive–language, social–personal, physiological, and overalldevelopment. Labyak and Metzger (1997) examined nine studiesthat sought to measure the effect of effleurage back massage onphysiological indicators of relaxation, and concluded that this formof MT was effective in promoting relaxation. However, interpre-tation of this finding is made problematic by their decision toinclude within-groups designs in the analysis, leaving open thepossibility that the observed effects could be attributable to spon-taneous recovery, placebo effect, or statistical regression (Field,1998, p. 1270), and by the fact that only limited information isprovided on the individual studies and their effect sizes. Ernst(1998) reviewed seven studies that assessed the effect of postex-ercise MT as a treatment for delayed-onset muscle soreness, reach-ing the tentative conclusion that MT may be a promising treatment,a conclusion that is hampered, like that of Labyak and Metzger, bya lack of sufficient statistics reported in the review itself.

No study to date has quantitatively reviewed the range ofcommonly reported MT effects in physically mature individuals.The present study is intended to address this problem. By means ofa more exhaustive literature search than those conducted in previ-ous reviews, we seek to unite the spectrum of MT studies thatappear in a range of scientific disciplines including psychology,medicine, nursing, and kinesiology. In addition, by limiting inclu-sion to studies that use a between-groups design with randomassignment of participants, the present study more accurately mea-sures MT’s true effects than reviews that have included otherdesigns that are open to bias and do not permit strong causalclaims.

Overview of MT

In modern practice, MT is not a single technique, or even asingle set of techniques. Rather, it is a broad heading for a rangeof approaches that share common characteristics, a fact that isevident in definitions provided by the AMTA. The AMTA definesmassage as “manual soft tissue manipulation [that] includes hold-

ing, causing movement, and/or applying pressure to the body,” andmassage therapy as “a profession in which the practitioner appliesmanual techniques, and may apply adjunctive therapies, with theintention of positively affecting the health and well-being of theclient” (AMTA, 1999a). Clearly, these definitions provide latitudefor a variety of approaches to exist under the rubric of MT. In oneinstance, MT may consist of a treatment lasting an hour or more,with long, firm strokes applied to numerous sites of the client’sbody, while that client lies partially disrobed on a specially de-signed table in a private clinic. In another instance, an MT clientmay receive a 10-min treatment of kneading focused on the shoul-ders while seated fully clothed in a specially designed chair, in apublic space such as a shopping mall or workplace. Duration oftreatment, types of touch and strokes administered, the sites of thebody where treatment is applied, the apparatus used to facilitatetreatment, and where that treatment takes place can all vary con-siderably. In addition, there is also considerable variability in theexplanatory mechanisms that massage therapists (and recipients)subscribe to. Finally, the outcomes being pursued may varywidely; whereas one client may undergo MT in the hopes ofobtaining relief from backache, another may receive MT to reduceemotional tension. In the present study, we define MT as themanual manipulation of soft tissue intended to promote health andwell-being, a definition that encompasses the diverse nature of thisform of treatment.

Though MT can take a variety of forms, the common elementthat allows these forms to be grouped together is their use ofinterpersonal touch in the form of soft tissue manipulation. Thiselement forms the basis for the predominant theories encounteredin MT research that are concerned with how it may provide thebenefits of reductions in anxiety, depression, stress hormones, andpain. In several of these theories, the pressure applied to the bodyby means of MT is thought to trigger certain physiological re-sponses that ultimately result in beneficial outcomes. It should benoted, however, that the pressure required by these theories has notbeen quantified, nor do existing clinical studies of MT routinelyreport on the amount of pressure administered in a way that wouldpermit precise replication. Although at least one study utilizinginfants as subjects observed differential effects in terms of weightgain for firm versus light strokes (Scafidi et al., 1986), no study todate has examined pressure as an independent variable with asample of physically mature participants.

MT Theories

Unfortunately, there has been little emphasis on theory in theMT literature, with many researchers choosing to emphasize theirpredictions and results without testing, or in some cases evendiscussing, possible explanatory mechanisms. In other instances,theories are offered, but important details are omitted. Researchershave rarely specified such things as whether a theory explainsimmediate versus lasting effects, or if activation of a theoreticalmechanism requires a course of treatment as opposed to a singleapplication. For the theories that follow, we suggest that only thefirst one, the gate control theory of pain reduction, is logicallylimited to providing an immediate effect. Each of the remainingtheories, to various degrees, could potentially offer immediate orlasting effects, or provide benefits that accumulate over a course of

4 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

treatment. However, it must be noted that these are strictly sup-positions and have not yet been tested.

The order in which these theories are presented reflects theirfrequency in the literature. Those that appear first are most fre-quently cited.

Gate Control Theory of Pain Reduction

Melzack and Wall (1965) theorized that the experience of paincan be reduced by competing stimuli such as pressure or cold,because of the fact that these stimuli travel along faster nervoussystem pathways than pain. In this way, MT performed withsufficient pressure would create a stimulus that interferes with thetransmission of the pain stimuli to the brain, effectively “closingthe gate” to the reception of pain before it can be processed (e.g.,Barbour, McGuire, & Kirchhoff, 1986; Field, 1998; Malkin,1994). This notion, that MT may have an analgesic effect consis-tent with gate control theory, appears in the literature more thanany other theory pertaining to MT.

Promotion of Parasympathetic Activity

MT may provide its benefits by shifting the autonomic nervoussystem (ANS) from a state of sympathetic response to a state ofparasympathetic response. A sympathetic response of the ANSoccurs as an individual’s body prepares to mobilize or defend itselfwhen faced with a threat or challenge, and is associated withincreased cardiovascular activity, an increase in stress hormones,and feelings of tension. Conversely, the parasympathetic responseoccurs when an individual’s body is at rest and not faced with athreat, or is recovering from a threat that has since passed, and isassociated with decreased cardiovascular activity, a decrease instress hormones, and feelings of calmness and well-being(Sarafino, 2002, p. 40).

The pressure applied during MT may stimulate vagal activity(Field, 1998, pp. 1273, 1276–1277), which in turn leads to areduction of stress hormones and physiological arousal, and asubsequent parasympathetic response of the ANS (e.g., Ferrell-Torry & Glick, 1993; Hulme, Waterman, & Hillier, 1999;Schachner, Field, Hernandez-Reif, Duarte, & Krasnegor, 1998).By stimulating a parasympathetic response through physiologicalmeans, MT may promote reductions in anxiety, depression, andpain that are consistent with a state of calmness. This samemechanism may also be responsible for several condition-specificbenefits resulting from MT, such as increased immune systemresponse in HIV-positive individuals (Diego et al., 2001), or im-proved functioning during a test of mental performance, in whichstudy participants receiving MT also displayed changes in electro-encephalograph pattern consistent with increased relaxation andalertness (Field, Ironson, et al., 1996). However, support for thistheory is not universal, and it has even been suggested that MTmay promote a sympathetic response of the ANS (e.g., Barr &Taslitz, 1970).

Influence on Body Chemistry

Two studies have linked MT with increased levels of serotonin(Field, Grizzle, Scafidi, & Schanberg, 1996; Ironson et al., 1996),

which “may inhibit the transmission of noxious nerve signals tothe brain” (Field, 1998, p. 1274). Others have suggested thatmanipulations such as rubbing, or applying pressure, may stimu-late a release of endorphins into the bloodstream (Andersson &Lundeberg, 1995; Oumeish, 1998). In these ways, MT may pro-vide pain relief or feelings of well-being by influencing the bodychemistry of the recipient.

Mechanical Effects

Articles concerned with sports performance, exercise recovery,and injury management highlight the possibility that MT mayspeed healing and reduce pain by mechanical means. The manip-ulations and pressure of MT may break down subcutaneous adhe-sions and prevent fibrosis (Donnelly & Wilton, 2002, p. 5) andpromote circulation of blood and lymph (Fritz, 2000, pp. 475–478), processes that may lead to reductions in pain associated withinjury or strenuous exercise. However, as a group, studies con-cerned with measuring MT’s effect on circulation have generatedinconsistent results (Tiidus, 1999).

Promotion of Restorative Sleep

Individuals deprived of deep sleep may experience changes inbody chemistry that lead to increases in pain. In the absence ofdeep sleep, levels of substance P increase and levels of somatosta-tin decrease, and both of these changes have been linked with theexperience of pain (Sunshine et al., 1996). Sunshine et al. (1996)concluded that MT may have promoted deeper, less disturbedsleep in a sample of fibromyalgia sufferers who experienced areduction in pain during the course of treatment. Chen, Lin, Wu,and Lin (1999) reached the conclusion that acupressure treatmentmay have been effective in improving sleep quality in a sample ofelderly residents at an assisted-living facility. In this way, MT mayreduce pain indirectly by promoting restorative sleep.

Interpersonal Attention

The five theories previously described, the majority of whichattempt to explain the role MT may play in reducing pain, are theonly ones that appear consistently in the scientific literature. How-ever, the element of interpersonal attention that may be present inMT must also be considered. It is occasionally noted that someportion of MT effects may result from the interpersonal attentionthat the recipient experiences, as opposed to resulting entirely fromthe activation of physiological mechanisms (Field, 1998, p. 1270;Malkin, 1994). However, although this possible effect of interper-sonal attention is acknowledged in the research literature, it isalmost universally treated as a nuisance variable, and comparisontreatments are selected in such a way that different groups receivethe same amount of attention. In this way it is believed that anybenefits demonstrated by the MT group that exceed those of thecomparison group can be attributed to a specific ingredient of MT,specifically interpersonal touch in the form of soft tissue manip-ulation. Although many studies, including all of those in thepresent analysis, attempt to control for interpersonal attention, nostudy to date has examined it as an independent variable. As such,

5MASSAGE THERAPY META-ANALYSIS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

the role that interpersonal attention may play in MT effects is notwell understood.

Effects

The present study examines both psychological and physiolog-ical effects resulting from MT. The psychological effects corre-spond with those suggested by Field and Gordon and endorsed bythe AMTA, and are also of interest because MT can be considereda novel way of treating these conditions, which are more routinelyaddressed by means of psychotherapy or pharmaceuticals. Thephysiological effects nominate themselves because MT is a phys-ical therapy.

We contend that MT effects can also be divided into single-doseeffects and multiple-dose effects. Single-dose effects include MT’sinfluence on states, either psychological or physiological, that aretransient in nature and that might reasonably be expected to beinfluenced by a single session of MT. These include state anxiety,negative mood, pain assessed immediately following treatment,heart rate, blood pressure, and cortisol level. Multiple-dose effectsare restricted to MT’s influence on variables that are typicallyconsidered to be more enduring, or that would likely be influencedonly by a series of MT sessions performed over a period of time,as opposed to a single dose. These variables include trait anxietyand depression, as well as pain when it is assessed at a timeconsiderably after treatment has ended.

Frequently, researchers elect to examine both single-dose effectsand multiple-dose effects within the same study. Diego et al.(2001) is one such study, in which treatment group participantsreceived MT twice weekly for a period of 12 weeks, and compar-ison group participants engaged in progressive muscle relaxation(PMR) according to the same schedule. Assessments of stateanxiety were made immediately prior to, and immediately follow-ing, both the first and last sessions of MT or PMR in the study.Depression, a condition expected to be more resistant to change,was assessed prior to the first session of MT or PMR, and not againuntil after the 24th and last sessions of either treatment. Manystudies, particularly those conducted by the Touch Research Insti-tute, use such a design in order to examine both single- andmultiple-dose effects.

It must be noted that the terms single-dose effect and multiple-dose effect are not yet in common usage. Research into MTgenerated by the Touch Research Institute typically uses the termsshort-term effect and long-term effect to make a similar distinction,but no consistent terminology has been used among other MTresearchers. The decision to use this terminology is motivated bythe desire to prevent any confusion that may arise with regard tohow long an effect may last following the termination of treatment.Very few studies have attempted to examine whether any MTeffects may last beyond the final day on which a participantreceives treatment, making the use of the term long-term effectpotentially confusing. All effects in the present study, with theexception of one outcome variable, were assessed on the same daythat a treatment took place. The exception is MT’s effect ondelayed assessment of pain, for which assessments took place atvarious time periods significantly after treatment had been discon-tinued. Presently, pain appears to be the only variable in the MTliterature that has been assessed in this way; the possibility that

MT may have enduring effects on other variables has gone essen-tially unaddressed.

Single-Dose Effects

State anxiety. State anxiety is a momentary emotional reactionconsisting of apprehension, tension, worry, and heightened ANSactivity. Because state anxiety can be understood as a reaction toone’s condition or environment, the intensity and duration of sucha state is determined by an individual’s perception of a situation asthreatening (Spielberger, 1972, p. 489). Many of the samples usedin MT research are drawn from populations experiencing seriousand chronic health problems that can lead to feelings of anxiety(Hughes, 1987; Popkin, Callies, Lentz, Cohen & Sutherland,1988). If MT is effective in reducing state anxiety, it may bedoubly valuable to such patient populations, in that it could bothimprove subjective well-being and promote physical health. Inphysically healthy populations, the improvement in subjectivewell-being alone may be the primary benefit of a reduction in stateanxiety.

Negative mood. Some studies have examined the effect of MTon mood, which may be defined as “transient episodes of feelingor affect” (Watson, 2000, p. 4). Although the primary studies donot specify a model for mood, virtually all the studies appear to beconcerned with MT’s ability to bring about a reduction of negativeaffect rather than an increase in positive affect.

Pain. Several studies have examined MT’s immediate effecton pain, the unpleasant emotional and sensory experience that isassociated with actual or potential tissue damage (Merskey et al.,1979). The sources of pain in the primary studies are diverse, andinclude conditions such as headache (Hernandez-Reif, Dieter,Field, Swerdlow, & Diego, 1998), backache (Hernandez-Reif,Field, Krasnegor, & Theakston, 2001), and labor pain (Hemenway,1993) among others.

Cortisol. Some MT studies have attempted to measure achange in participants’ cortisol levels. Cortisol is a stress hormoneassociated with the sympathetic response of the ANS (Field,1998). MT, a therapy commonly thought of as relaxing, is ex-pected to reduce cortisol levels, a finding that would be consistentwith facilitating a parasympathetic response of the ANS (e.g.,Field et al., 1992; Ironson et al., 1996).

Blood pressure. A handful of studies have examined MT’seffect on blood pressure. Although predictions are not alwaysoffered, most commonly MT is expected to reduce blood pressureconsistent with a parasympathetic response of the ANS(Hernandez-Reif, Field, et al., 2000; Okvat, Oz, Ting, &Namerow, 2002).

Heart rate. A few studies examining MT have attempted tomeasure its physiological effects in terms of heart rate. Research-ers have not always offered clear predictions for this variable (Barr& Taslitz, 1970), but in cases where a prediction is evident, mostoften a decrease in heart rate is predicted, consistent with aparasympathetic response of the ANS (Cottingham, Porges, &Richmond, 1988; Okvat et al., 2002). Nevertheless, some research-ers have noted that the opposite effect could be observed in casesin which MT was a novel experience for research participants(Reed & Held, 1988, p. 1232).

6 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Multiple-Dose Effects

Trait anxiety. Several studies have examined MT’s potentialto reduce trait anxiety, the “relatively stable individual differencesin anxiety proneness as a personality trait” (Spielberger, 1972, p.482). In contrast with the transient and situation-specific nature ofstate anxiety, trait anxiety is a dispositional, internalized pronenessto be anxious (Phillips, Martin, & Meyers, 1972, p. 412). Personswith high levels of trait anxiety tend to perceive the world as moredangerous or threatening, and experience anxiety states more fre-quently and with greater intensity than those with lower levels oftrait anxiety (Spielberger, 1972, p. 482).

Depression. Ingram and Siegle (2002) noted that, in the courseof research, the concept of depression has been defined manydifferent ways, including as a mood state, a symptom, a syndrome,a mood disorder, and a disease. In the current meta-analysis,studies included in this category have been chosen on the basis oftheir utilization of a measure believed to capture something be-yond “ordinary unhappiness” or a “sad mood,” symptoms thatwould more accurately belong to the previously discussed categoryof negative mood. Subclinical depression, likely the best descrip-tion of the type of depression most often assessed in MT research,consists of the aforementioned symptoms combined with symp-toms such as mild to moderate levels of motivational and cognitivedeficits, vegetative signs, and disruptions in interpersonal relation-ships (Ingram & Siegle, 2002, p. 90).

Delayed assessment of pain. A few studies have assessedparticipants’ experience of pain at one or more time points signif-icantly after a course of treatment has ended. The majority of thesestudies have done so at intervals that range from a few days to 6weeks (Cen, 2000; Dyson-Hudson, Shiflett, Kirshblum, Bowen, &Druin, 2001; Preyde, 2000; Shulman & Jones, 1996), although onestudy included an assessment that took place 42 weeks aftertreatment ended (Cherkin et al., 2001). Because of the smallnumber of studies, and the range of times at which delayedassessments were made, it is not expected that the present studywill be able to determine precisely how long an analgesic effectresulting from MT lasts, or the rate at which such an effect decays;rather, the aim is simply to examine whether or not MT may havea lasting analgesic effect.

Moderators

A number of potentially interesting moderator variables havegone unexamined in MT research. Primary studies, for instance,have neglected to examine whether the length of MT sessions, orcharacteristics of the therapist and the recipient, influence themagnitude of MT effects. Similarly, only a few studies have usedmore than one comparison group, making it difficult to determinewhether the type of treatment to which MT is compared maymoderate its effects. Although within-study examinations of suchmoderators would permit stronger inferences to be made, theirimportance can be explored in the present study by means ofbetween-study comparisons. In addition, the present study alsoexamines a potential moderator that cannot be examined within anindividual study, that of a laboratory effect.

Minutes of MT per session. It is common for treatment studiesin medicine (e.g., Bollini, Pampallona, Tibaldi, Kupelnick, &

Munizza, 1999; Yyldyz & Sachs, 2001) and in psychotherapy(e.g., Bierenbaum, Nichols, & Schwartz, 1976; Turner, Valtierra,Talken, Miller, & DeAnda, 1996) to examine dosage as an inde-pendent variable. However, no studies concerned with MT havedone so. It is not known whether there is a minimal amount, interms of minutes of MT administered per session, required toproduce benefits, nor is it known whether there is an optimalamount of MT that produces benefits most efficiently. Fortunately,the studies that exist vary considerably in the amount of MTadministered to participants in each session, from as little as 5 min(Fraser & Kerr, 1993; Wendler, 1999) to as much as an hour(Levin, 1990). By examining the relationship between the magni-tude of effects generated and the amount of MT administered persession, the present study aims to determine whether there areminimum or optimum dosages of MT.

Mean age of participants. Although MT research has beenperformed on samples with a variety of age ranges, no study hassought to determine whether MT offers effects of differing mag-nitude to participants who differ in age. The present study exam-ines whether there is a relationship between the mean age of theparticipants in a study and the magnitude of effects.

Gender of participants. Only one study to date, using a verysmall sample, has examined whether MT effects might vary ac-cording to the gender of the recipients (Weinrich & Weinrich,1990). The present study more powerfully examines the possibilitythat the gender of the recipient might moderate MT effects byexamining whether study outcomes vary according to gender.

Type of comparison treatment. In discussing the research find-ings for a different treatment modality (psychotherapy), Wampold(2001) noted that there is a distinction that must be made betweenabsolute and relative efficacy. Absolute efficacy “refers to theeffects of treatment vis-a-vis no treatment and accordingly is bestaddressed by a research design where treated participants arecontrasted with untreated participants” (Wampold, 2001, p. 59).By contrast, relative efficacy “is typically investigated by compar-ing the outcomes of two treatments” when one wishes to determinewhich, if either, is superior (Wampold, 2001, p. 73). Clearly, thetype of efficacy one wishes to measure plays an important part indetermining what will be an appropriate choice for a comparison,as a study designed to measure one does not necessarily measurethe other. This issue of distinguishing absolute efficacy (does MTwork better than no treatment at all?) from relative efficacy (doesMT work better than a specific alternative treatment, such asPMR?) has not been made explicit enough in MT research. How-ever, a wide variety of comparison treatments have been used inMT research, some of which resemble a wait-list (no treatment)condition, whereas others use active treatments (such as the afore-mentioned PMR, or chiropractic care) as a point of comparison, orplacebo-type comparison treatments that are meant to account forthe effect of receiving attention (such as transcutaneous electricalstimulation performed with a machine that is not delivering anycurrent to the participant). Logically, if MT has any effect what-soever, we expect the MT effects that result from comparison witha no-treatment condition would be larger than those that resultfrom comparing MT to any treatment condition, including so-called placebo conditions in which the participants receive noviable treatment. Combining the results of such different studieswithout attempting to account for these different comparison

7MASSAGE THERAPY META-ANALYSIS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

points could be problematic. For this reason, we have divided thecomparison treatments in the primary studies, when possible, asbelonging to either wait-list equivalent or active/placebocategories.

The wait-list equivalent category consists of comparison treat-ments that most closely resemble having received no treatment,and includes wait-list controls, standard care (in studies where allparticipants had a medical condition and continued to receive carefor that condition regardless of group assignment), rest, reading, ora work break. The active/placebo category consists of all othercomparison treatments, which are grouped according to the expec-tation that each could reasonably be expected to have some effect,including the possibility of a placebo effect. These include treat-ments such as PMR, acupuncture, chiropractic care, and variousforms of attention, among others. Studies that used multiple com-parison groups that could not be included together within a singlecategory were not included in either category.

Therapist training. Treatment research in fields such as psy-chology (Pinquart & Soerensen, 2001; Weisz, Weiss, Alicke, &Klotz, 1987) and medicine (Lin et al., 1997; Tiemens et al., 1999)sometimes examines the existence of training effects to determinewhether practitioners with greater amounts of training providegreater benefit to those being treated. No MT research, however,has examined the training of the massage therapist as an indepen-dent variable. However, the studies that do exist vary in regard towho performs MT on participants. The majority of studies use oneor more fully trained and licensed massage therapists. Othersutilize a layperson with only minimal training in providing mas-sage, usually just enough to facilitate the study (e.g., Fischer,Bianculli, Sehdev, & Hediger, 2000; Weinrich & Weinrich, 1990;Wendler, 1999). By contrasting the results of studies that used afully trained massage therapist with those that used a layperson toprovide treatment, the present meta-analysis may be able to deter-mine whether a therapist’s training plays an important role inproviding MT benefits.

Laboratory effect. Much of the research in this area, andespecially the most recent research, is the product of a singlelaboratory, the Touch Research Institute (Field, 1998). Becausethis one source is responsible for a large proportion of MT studies,it is important to determine whether the results coming from thisresearch group differ in a significant way from those of otherresearchers. If a difference is found, it would be important toexamine more closely what factors contribute to that difference.

Predictions

MT is expected to promote significant and desirable reductionsfor each of the following variables, consistent with the existingexplanatory theories outlined above: state anxiety, negative mood,pain (immediate and delayed assessment), cortisol, heart rate,blood pressure, trait anxiety, and depression. It is expected thatgreater reductions in these variables will be associated with higherdoses of MT, in the form of minutes of MT administered persession, a relationship one would expect to observe if MT is aviable treatment. MT effects are not expected to vary according tothe age or gender of participants. It is expected that MT effectsgenerated from studies using wait-list equivalent comparison treat-ments will be larger than those generated from studies with active/

placebo comparison treatments. Finally, no prediction is madeconcerning therapist training, or the existence of a laboratoryeffect.

Method

Literature Search and Criteria for Inclusion

A literature search was performed by Christopher A. Moyer and agraduate student in library and information sciences hired as a researchassistant. The PsycINFO, MEDLINE, CINAHL, SPORT Discus, and Dis-sertation Abstracts International databases were searched using the fol-lowing key words: massage, massotherapy, acupressure (and accupres-sure), applied kinesiology, bodywork, musculoskeletal manipulation,reflexology, relaxation techniques, Rolfing, Touch Research Institute, andTrager. Author searches were conducted within the same databases for thefollowing authors associated with MT research: Burman, I.; Field, T.; Hart,S.; Hernandez-Reif, M.; Kuhn, C.; Peck, M.; Quintino, O.; Schanberg, S.;Taylor, S.; Theakston, H.; Weinrich, M.; and Weinrich, S. The InternetWeb sites of the AMTA (www.amtamassage.org), the AMTA Foundation(www.amtafoundation.org), and the Touch Research Institute (http://www.miami.edu/touch-research/) were inspected for references, and the TouchResearch Institute was also contacted directly to request unpublished data.The reference lists of all studies located by these means were then manu-ally searched to yield additional studies.

All studies were inspected to ensure that they examined a form of MTconsistent with the present study’s operational definition, in which MT isdefined as the manual manipulation of soft tissue intended to promotehealth and well-being. Studies were limited to those that administered MTto human participants other than infants, and that reported results inEnglish. Studies concerned with chiropractic, heat therapy, hydrotherapy,passive motion, or progressive relaxation treatments were not included,unless the study also included an MT group. Studies examining therapeutictouch, a nursing intervention distinct from MT (in that it does not actuallyrequire physical contact to occur), were also excluded unless they also hadan MT group. Several studies used more than two groups; in these cases,study results were combined in order to yield a between-groups compari-son of all subjects receiving MT versus all subjects receiving non-MTtreatments. Studies concerned with ice massage, participants performingself-massage, or massage performed with the aid of mechanical deviceswere excluded, as were studies that only included MT as part of acombination treatment (e.g., MT combined with exercise and movementtherapy). MT administered with scented oil or MT administered withbackground music were not considered to be combination treatments, asthese are common elements of MT in clinical practice, and studies usingsuch treatment were included. Studies that did not explicitly label atreatment as “massage” or as “massage therapy,” but used a treatment thatfit the authors’ operational definition of MT, were included.

These criteria yielded 144 studies concerned with outcomes of MT. Eachstudy was reviewed independently by Christopher A. Moyer and JamesRounds for possible inclusion in the meta-analysis. Studies were examinedto ensure that they (a) compared an MT group with one or more non-MTcontrol groups, (b) used random assignment to groups, and (c) reportedsufficient data for a between-groups effect size to be generated on at leastone dependent variable of interest. These three criteria accounted forapproximately equal proportions of excluded studies.

The first two inclusion criteria were necessary to ensure that effects werea result of treatment. When participants in MT research serve as their owncontrols (e.g., Bauer & Dracup, 1987; Fakouri & Jones, 1987) there is noway to know whether effects are attributable to treatment or are instead theresult of spontaneous recovery, placebo effect, or statistical regression(Field, 1998, p. 1270). Similarly, random assignment of participants togroups is necessary to control for the possibility of selection effects. Glaser

8 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

(1990) is an example of a study that is threatened in this way. Becausetreatment participants were previously enrolled in an MT program, andwere compared with a group of participants who were not enrolled, it islikely that these groups differed in their predisposition toward MT in a waythat could affect results.

When studies met all criteria apart from reporting sufficient data forcalculating between-groups effects, and contact information was available,study authors were contacted in an attempt to obtain the necessary data.Specifically, there were seven studies from the Touch Research Institutefor which this was the case (Field et al., 1999; Field et al., 2000; Field,Peck, et al., 1998; Field, Quintino, Henteleff, Wells-Keife, & Delvecchio-Feinberg, 1997; Field, Schanberg, et al., 1998; Field, Sunshine, et al., 1997;Sunshine et al., 1996). Upon our request, we were informed that the dataneeded from these studies (standard deviations) were no longer available.For this reason, these studies could not be included in the meta-analysis.

Interrater agreement for the inclusion process was 93%. The 10 studiesfor which there was initial disagreement, which occurred most frequentlyas a result of uncertainty regarding random assignment, were then reviewedjointly, with the subsequent decision made to exclude 8 of these. Thisresulted in a total of 37 studies meeting the inclusion criteria.

Variables and MeasuresThe nine variables for which effect sizes were calculated, and the

instruments used to assess them, are as follows:State anxiety. Fifteen of the 21 studies examining MT’s effect on

anxiety used the state anxiety portion of the State–Trait Anxiety Inventory(Spielberger, 1983). Five studies used a visual analogue scale, and onestudy used an investigator-constructed measure.

Negative mood. Seven of eight studies assessing negative mood usedthe Profile of Mood States (McNair, Lorr, & Droppleman, 1971). Theremaining study used a visual analogue scale.

Immediate assessment of pain. Eight of the 15 studies assessing painimmediately following treatment used visual analogue scales alone. Twostudies used a visual analogue scale in conjunction with either the Short-Form McGill Pain Questionnaire (Melzack, 1987) or the Menstrual Dis-tress Questionnaire (Moos, 1968). Two studies used investigator-constructed measures, and the remaining studies relied on the Neck PainQuestionnaire (Leak et al., 1994), the revised Oswestry Low Back PainQuestionnaire (Hudson-Cook, Tomes-Nicholson, & Breen, 1989), or be-havioral observation.

Cortisol. Of the seven studies that assessed cortisol levels, four reliedon salivary samples, two on urinary samples, and one on a blood sample.In each case, samples were collected 20 min after the application of MT,to account for the fact that bodily cortisol levels are indicative of responsesoccurring 20 min prior to sampling (Field, Hernandez-Reif, Quintino,Schanberg, & Kuhn, 1998, p. 233).

Blood pressure. Five studies offer data pertaining to participants’blood pressure, assessed by means of a sphygmomanometer. Measures ofdiastolic and systolic blood pressure were combined into one effect size,because only a few studies report on this variable, and differ in regard towhich values they report.

Heart rate. Of the six studies that assessed the effect of MT on heartrate, four used some type of automatic monitoring device, and one studyindicated that pulse was assessed manually. One study did not specify themeans by which heart rate was assessed.

Trait anxiety. Three studies of the seven assessing trait anxiety usedthe Symptom Checklist-90–Revised (SCL-90-R; Derogatis, 1983). Onestudy combined the Conners Teacher Rating Scale (Conners, 1969) and theRevised Children’s Manifest Anxiety Scale (Reynolds & Richmond,1985). The three remaining studies used either the Beck Anxiety Inventory(Beck, Brown, Epstein, & Steer, 1988), the trait portion of the State–TraitAnxiety Inventory (Spielberger, 1983), or an investigator-constructedmeasure.

Depression. Five of the 10 studies assessing depression utilized theCenter for Epidemiological Studies—Depression Scale (CES–D; Radloff,1977). Two used the SCL-90-R, and one combined the CES–D and theSCL-90-R. The remaining studies used either the Children’s DepressionInventory—Short Form (Kovacs, 1992) or an investigator-constructedmeasure.

Delayed assessment of pain. The five studies assessing pain at a timesignificantly after treatment ended relied on five different instruments.These were the Neck Pain Questionnaire (Leak et al., 1994), the Wheel-chair User’s Shoulder Pain Index (Curtis et al., 1995), the McGill PainQuestionnaire (Melzack, 1975), a visual analogue scale, and aninvestigator-constructed measure.

Statistical AnalysisEffect sizes. Between-groups comparisons on variables of interest were

converted to Hedges’s g effect size. Hedges’s g, calculated as (Group Mean1 – Group Mean 2) � pooled standard deviation, estimates the number ofstandard deviations by which the average member of a treatment groupdiffers from the average member of a comparison group for a givenoutcome. In cases where a study used more than one measure to examinethe same outcome variable, results of multiple measures were standardizedand then averaged in order to result in one effect size per variable for anystudy. Similarly, if a study examined the immediate effects of more thanone application of treatment, or examined the treatment effect on delayedassessments of pain at more than one time point, the results of the multipleapplications or assessments were standardized and then averaged in orderto calculate a single effect size for that study. Effect sizes were coded suchthat positive values, for any variable, indicate a more desirable outcome(e.g., a reduction in anxiety) for the participants who received MT.

This process was done independently by both the first and secondauthors for the entire set of effect sizes; these initial results were thencompared in order to determine agreement and eliminate errors. Agreementrate (AR) of initial calculations for the entire set of 84 effect sizes was 88%.Within outcome categories, the initial rates of agreement were as follows:state anxiety, AR � 86% (n � 21); negative mood, AR � 88% (n � 8);immediate assessment of pain, AR � 87% (n � 15); cortisol, AR � 86%(n � 7); blood pressure, AR � 60% (n � 5); heart rate, AR � 100% (n �6); trait anxiety, AR � 86% (n � 7); depression, AR � 90% (n � 10); anddelayed assessment of pain, AR � 60% (n � 5). When discrepancies wereobserved, calculations were reviewed jointly to correct errors, and a con-sensus was reached.

Individual study effect sizes were then subjected to a correction for smallsample bias, then weighted by their inverse variance and averaged togenerate a mean effect size for each outcome variable (Lipsey & Wilson,2001). An overall, nonspecific effect size was also calculated by averagingall effects within each study, and then calculating a weighted overall effectfrom these effect sizes. All effect sizes were calculated according to arandom effects model of error estimation.

Statistical significance of the mean effect sizes was assessed by calcu-lating the 95% confidence interval (CI) for the population parameter. Asignificance level of .05 or better is inferred when zero is not containedwithin the CI. For effect sizes reaching statistical significance, the likeli-hood and possible influence of publication bias—the possibility that stud-ies retrieved for the meta-analysis may not be a random sample of allstudies actually conducted (Rosenthal, 1998)—was assessed by means of atrim and fill procedure (Duval & Tweedie, 2000), a nonparametric statis-tical technique of examining the symmetry and distribution of effect sizesplotted by inverse variance. This technique first estimates the number ofstudies that may be missing as a result of publication bias, and then allowsa new, attenuated effect size to be calculated on the basis of the influencesuch studies would have if they were included in the analysis. The trim andfill procedure was performed with the Division of Vector-Borne InfectiousDiseases library using the statistical computing program S-PLUS (Bigger-

9MASSAGE THERAPY META-ANALYSIS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Table 1Individual Study Characteristics and Effect Sizes (g) by Outcome Variable

Study Participants N % femaleMeanage

Min/session

Comp.type

Trainedtherapist?

TRIstudy? g

State anxiety

Chang et al. (2002) Pregnant women 60 100 28 30 WL No No 0.45Chin (1999) Surgery patients 85 100 42 10 WL No No �0.50Delaney et al. (2002) Healthy adults 30 53 31 20 WL Yes No 0.20Diego et al. (2002) Spinal cord patients 20 25 39 40 A/P Yes Yes 0.57Diego et al. (2001) HIV� adolescents 24 92 17 20 A/P Yes Yes 0.87Field et al. (2002) Fibromyalgia patients 20 — 51 30 A/P Yes Yes 0.11Field, Ironson, et al. (1996) Medical staff 50 80 26 15 A/P Yes Yes 0.48Fischer et al. (2000) Amniocentesis patients 200 100 34 — WL No No 0.00Fraser & Kerr (1993) Institutionalized elderly 21 — — 5 C — No 1.20Groer et al. (1994) Healthy adults 32 69 64 10 WL No No �0.21Hernandez-Reif, Field, et al. (1998) Multiple sclerosis patients 24 75 48 45 WL Yes Yes 1.33Hernandez-Reif et al. (2001) Back pain patients 24 54 40 30 A/P Yes Yes 0.07Hernandez-Reif, Field, et al. (2000) Hypertensive adults 30 53 52 30 A/P Yes Yes 0.24Hernandez-Reif, Martinez, et al. (2000) PDD patients 22 100 33 30 A/P Yes Yes 0.84Leivadi et al. (1999) University dance students 30 100 20 30 A/P Yes Yes 0.21Levin (1990) Healthy adults 36 — 27 60 WL Yes No 1.30Menard (1995) Surgery patients 30 100 52 45 WL Yes No 1.12Mueller Hinze (1988) Healthy women 48 100 27 10 C — No 0.50Okvat et al. (2002) Cardiac catheter patients 78 24 61 10 A/P Yes No �0.06Richards (1993) Hospitalized elderly men 69 0 66 6 C No No 0.80Wendler (1999) Soldiers 93 10 30 5 A/P No No 0.54

Negative mood

Abrams (1999) Children/adolescents with ADHD 30 17 13 20 WL Yes Yes 0.09Field et al. (2002) Fibromyalgia patients 20 — 51 30 A/P Yes Yes 0.00Field, Ironson, et al. (1996) Medical staff 50 80 26 15 A/P Yes Yes 1.09Hernandez-Reif, Field, et al. (1998) Multiple sclerosis patients 24 75 48 45 WL Yes Yes 0.32Hernandez-Reif et al. (2001) Back pain patients 24 54 40 30 A/P Yes Yes �0.07Hernandez-Reif, Martinez, et al. (2000) PDD patients 24 100 33 30 A/P — Yes 1.27Leivadi et al. (1999) University dance students 30 100 20 30 A/P Yes Yes �0.49Levin (1990) Healthy adults 36 — 27 60 WL Yes No 0.46

Immediate assessment of pain

Cen (2000) Neck pain patients 31 75 48 30 C Yes No 1.21Chang et al. (2002) Pregnant women 60 100 28 30 WL No No 0.99Chin (1999) Surgery patients 85 100 42 10 WL No No �0.30Field et al. (2002) Fibromyalgia patients 20 — 51 30 A/P Yes Yes 0.85Fischer et al. (2000) Amniocentesis patients 200 100 34 — WL No No �0.13Hemenway (1993) Labor pain patients 32 100 23 10 A/P No No 0.38Hernandez-Reif, Dieter, et al. (1998) Headache patients 26 — 40 30 WL Yes Yes 0.52Hernandez-Reif et al. (2001) Back pain patients 24 54 40 30 A/P Yes Yes 0.35Hernandez-Reif, Martinez, et al. (2000) PDD patients 24 100 33 30 A/P — Yes 0.81Hsieh et al. (1992) Back pain patients 63 — 34 — A/P Yes No �0.94Leivadi et al. (1999) University dance students 30 100 20 30 A/P Yes Yes 0.21Mueller Hinze (1988) Healthy women 48 100 27 10 C — No 0.81Okvat et al. (2002) Cardiac catheter patients 78 24 61 10 A/P Yes No 0.16Weinrich & Weinrich (1990) Cancer patients 28 36 62 10 A/P No No �0.04Wilkie et al. (2000) Hospice care cancer patients 29 31 63 30 WL Yes No �0.14

Cortisol

Abrams (1999) Children/adolescents with ADHD 30 17 13 20 WL Yes Yes 0.07Chin (1999) Surgery patients 85 100 42 10 WL No No 0.07Field, Ironson, et al. (1996) Medical staff 50 80 26 15 A/P Yes Yes 0.45Hernandez-Reif et al. (2001) Back pain patients 24 54 40 30 A/P Yes Yes �0.39Hernandez-Reif, Field, et al. (2000) Hypertensive adults 30 53 52 30 A/P Yes Yes 0.18Hernandez-Reif et al. (2002) Parkinson’s disease patients 16 50 58 30 A/P Yes Yes 0.41Leivadi et al. (1999) University dance students 30 100 20 30 A/P Yes Yes 0.13

10 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

staff, 2000), which generates results for the three estimators of missingstudies (L0, R0, and Q0) described by Duval and Tweedie (2000). Per thesuggestion of these authors, the number of missing studies resulting fromeach estimator was considered before the eventual decision was made toreport results according to the L0 and R0 estimators, which are consideredpreferable for most situations (Duval & Tweedie, 2000).

Moderators. As with effect sizes, moderator variable data were alsocoded independently by both the first and second authors. Agreement ratefor initial coding of all moderator data across categories was 97% (n �158). Within moderator variable categories, initial agreement rates were asfollows: minutes per session, AR � 100% (n � 34); mean age, AR � 100%(n � 25); comparison type, AR � 97% (n � 34); training, AR � 87% (n �

31); and laboratory effect, AR � 100% (n � 34); proportion of femaleparticipants was coded only by the first author. The influence of moderatorvariables was assessed by performing a weighted regression analysis(Lipsey & Wilson, 2001) on the set of overall, nonspecific effect sizes forall studies.

Results

Table 1 lists the effect sizes (Hedges’s g) for each study byoutcome variable, as well as important study characteristics. The37 studies included in the meta-analysis used a total of 1,802

Table 1 (continued)

Study Participants N % femaleMeanage

Min/session

Comp.type

Trainedtherapist?

TRIstudy? g

Blood pressure

Delaney et al. (2002) Healthy adults 30 53 31 20 WL Yes No �0.06Hernandez-Reif, Field, et al. (2000) Hypertensive adults 30 53 52 30 A/P Yes Yes 0.29Mueller Hinze (1988) Healthy women 48 100 27 10 C — No 0.49Okvat et al. (2002) Cardiac catheter patients 78 24 61 10 A/P Yes No 0.16Wendler (1999) Soldiers 93 10 30 5 A/P No No 0.34

Heart rate

Cottingham et al. (1988) Healthy men 32 0 27 45 WL Yes No 0.22Delaney et al. (2002) Healthy adults 30 53 31 20 WL Yes No 0.53Mueller Hinze (1988) Healthy women 48 100 27 10 C — No 0.82Okvat et al. (2002) Cardiac catheter patients 78 24 61 10 A/P Yes No 0.16Richards (1993) Hospitalized elderly men 69 0 66 6 C No No 0.35Wendler (1999) Soldiers 93 10 30 5 A/P No No 0.52

Trait anxiety

Abrams (1999) Children/adolescents with ADHD 30 17 13 20 WL Yes Yes 0.94Hernandez-Reif, Dieter, et al. (1998) Headache patients 26 — 40 30 A/P Yes Yes 0.52Hernandez-Reif et al. (2001) Back pain patients 24 54 40 30 A/P Yes Yes 0.98Hernandez-Reif, Field, et al. (2000) Hypertensive adults 30 53 52 30 A/P Yes Yes 2.11Rexilius et al. (2002) Patient caregivers 35 72 52 30 C Yes No 0.31Scherder et al. (1998) Alzheimer’s patients 16 — 86 30 A/P — No 0.68Shulman & Jones (1996) Employees 33 61 40 15 WL Yes No 0.06

Depression

Abrams (1999) Children/adolescents with ADHD 30 17 13 20 WL Yes Yes 0.29Diego et al. (2002) Spinal cord patients 20 25 39 40 A/P Yes Yes 0.32Diego et al. (2001) HIV� adolescents 24 92 17 20 A/P Yes Yes 0.74Field et al. (2002) Fibromyalgia patients 20 — 51 30 A/P Yes Yes 0.63Hernandez-Reif, Dieter, et al. (1998) Headache patients 26 — 40 30 WL Yes Yes 0.38Hernandez-Reif et al. (2001) Back pain patients 24 54 40 30 A/P Yes Yes 0.80Hernandez-Reif, Field, et al. (2000) Hypertensive adults 30 53 52 30 A/P Yes Yes 0.82Hernandez-Reif, Martinez, et al. (2000) PDD patients 24 100 33 30 A/P — Yes 0.28Rexilius et al. (2002) Patient caregivers 35 72 52 30 C Yes No 0.91Scherder et al. (1998) Alzheimer’s patients 16 — 86 30 A/P — No 1.50

Delayed assessment of pain

Cen (2000) Neck pain patients 31 75 48 30 C Yes No 0.36Cherkin et al. (2001) Back pain patients 262 58 45 — C Yes No 0.25Dyson-Hudson et al. (2001) Wheelchair users 18 22 45 45 A/P Yes No 0.35Preyde (2000) Back pain patients 73 51 45 30 C Yes No 0.49Stratford et al. (1989) Tendinitis patients 40 50 43 10 WL — No 0.30

Note. Dashes indicate that data were not reported. Comp. � comparison; TRI � Touch Research Institute; A/P � active/placebo; C � combination; WL� wait-list equivalent; PDD � premenstrual dysphoric disorder; ADHD � attention-deficit/hyperactivity disorder.

11MASSAGE THERAPY META-ANALYSIS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

participants, including 795 who received MT. Of the 1,007 par-ticipants who received a comparison treatment, 49% received oneof the five treatments categorized as wait-list equivalent, and theremaining 51% received a treatment categorized as active/placebo.The mean number of participants for a study was 48.7 (SD �49.0), and mean age of all participants was 40.6 years (SD � 13.9).Participants received an average of 21.7 min (SD � 14.0) of MTper application of treatment. Sixty-five percent of studies reportedusing a trained massage therapist (or therapists), 22% reportedusing a minimally trained person (or persons) to deliver treatment,and 14% did not indicate the level of training of the person (orpersons) administering MT. Thirty-two percent of studies wereconducted by the Touch Research Institute.

Table 2 graphically represents the distribution of overall studyeffect sizes by means of a stem and leaf plot. Table 3 lists the meaneffect size for each outcome variable, as well as the number ofstudies contributing to the effect size, its 95% CI, and the resultsof trim and fill procedures applied to statistically significant ef-fects. The nonspecific, overall mean effect was statistically signif-icant (g � 0.34, p � .01). Among the nine specific outcomevariables examined, six displayed statistically significant effectsizes. For the single-dose effects category, these included stateanxiety (g � 0.37, p � .01), blood pressure (g � 0.25, p � .02),and heart rate (g � 0.41, p � .01). Negative mood (g � 0.34),immediate assessment of pain (g � 0.28) and cortisol (g � 0.14)were nonsignificant. All outcome variables examined within themultiple-dose effects category, including trait anxiety (g � 0.75,p � .01), depression (g � 0.62, p � .01), and delayed assessmentof pain (g � 0.31, p � .01), were statistically significant.

The results of trim and fill analyses conducted on the statisti-cally significant outcome variables indicated that the results arefairly robust to the threat of publication bias. For overall effects, an

analysis based on the L0 estimator yielded 10 studies missing as aresult of publication bias, which result in an attenuated but stillsignificant effect (g � 0.20, 95% CI � 0.06, 0.34); the funnel plotof actual and filled study effect sizes for this analysis is repre-sented in Figure 1. The same analysis performed with the R0estimator indicates no missing studies. Of the six specific outcomevariables that generated significant effects, results of trim and fillanalyses indicated that only state anxiety and delayed assessmentof pain effects were likely overestimated due to publication bias. Atrim and fill analysis performed on the state anxiety effect usingthe L0 estimator yielded an estimate of four studies likely missingas a result of publication bias. When the influence such studieswould have on state anxiety is calculated, the adjusted effect isnonsignificant (g � 0.22, 95% CI � �0.01, 0.45). A trim and fillanalysis performed on the delayed assessment of pain outcomevariable using the L0 estimator yielded a slightly smaller but stillsignificant effect (g � 0.26, 95% CI � 0.07, 0.44). When the sameanalyses were performed with the R0 estimator, no missing studieswere indicated in either case.

An analysis of potential moderator variables for the set ofoverall effect sizes was not statistically significant, QR(6) � 5.80.Despite the nonsignificance of the regression model, the decisionwas made to inspect the significance of the individual moderatorvariables. Minutes of MT administered per session (z � 1.55, p �.06, one-tailed) was the only moderator that approached the pre-determined alpha for statistical significance ( p � .05). To examinethis variable a bit further, we calculated separate weighted effectsizes for two categories of studies. Studies that administered � 30min of MT per session generated an effect that was substantiallylarger than that resulting from the entire set of studies (g � 0.54,95% CI � 0.32, 0.76). Studies that administered � 30 min of MTper session demonstrated an effect that was slightly smaller thanthat of the entire set of studies, but still significant (g � 0.30, 95%CI � 0.08, 0.52).

Discussion

This meta-analysis supports the general conclusion that MT iseffective. Thirty-seven studies yielded a statistically significantoverall effect as well as six specific effects out of nine that wereexamined. Significant results were found within the single-doseand multiple-dose categories, and for both physiological and psy-chological outcome variables. Confidence in these findings isbolstered by the results of trim and fill analyses, which indicatethat the results are not unduly threatened by publication bias.

Single-Dose Effects

Three of the six single-dose effects examined were statisticallysignificant. The magnitude of MT’s effect on state anxiety meansthat the average participant receiving MT experienced a reductionof state anxiety that was greater than 64% of participants receivinga comparison treatment. MT was also more effective than com-parison treatments in reducing blood pressure and heart rate. Theaverage MT participant experienced a reduction in blood pressurethat was greater than 60% of comparison group participants,whereas for heart rate, the reduction resulting from MT was greaterthan 66% of comparison group participants, findings that are

Table 2Stem and Leaf Plot of 37 Overall Study Effect Sizes

Stem Leaf

�0.9 4�0.8�0.7�0.6�0.5�0.4�0.3�0.2 14�0.1 4�0.0 47

0.0 260.1 10.2 22590.3 05580.4 01145790.5 80.6 170.7 23890.8 0130.91.0 91.1 21.2 0

12 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

consistent with the theory that MT may promote a parasympatheticresponse of the ANS. Cortisol, however, another outcome variablethat would be expected to decrease if MT promotes a parasympa-thetic response, was not significantly reduced, a finding that con-trasts with the conclusion previously reached by Field (1998).Despite this inconsistent support for MT promoting a parasympa-thetic response, the significant finding for the cardiovascular vari-ables suggests that future research should examine whether MTmight have an enduring effect on blood pressure such that it couldbe used in treating hypertension.

MT did not exhibit an effect on immediate assessment of pain.This finding contrasts with the commonly offered notion that MT

may provide analgesia by competing with painful stimuli in a wayconsistent with the gate control theory of pain. MT’s effect onnegative mood was also nonsignificant.

Multiple-Dose Effects

Some of MT’s largest and most interesting effects belong to themultiple-dose effects category. Despite the fact that MT did notdemonstrate an effect on immediate assessment of pain, a signif-icant effect was found for delayed assessment of pain. MT partic-ipants who received a course of treatment and were assessedseveral days or weeks after treatment ended exhibited levels of

Figure 1. Funnel plot of 37 overall study effect sizes (g) plus the 10 effect sizes filled in by means of trim andfill procedure using the L0 estimator; no filled studies are indicated using the R0 estimator.

Table 3Mean Effect Sizes (g) and Results of Trim and Fill Analyses by Outcome Variable

Outcome variable k g 95% CI L0

Adjusted g based onk � L0 Adjusted 95% CI

Overall 37 0.34** 0.21, 0.48 10 0.20** 0.06, 0.34Single-dose effects

State anxiety 21 0.37** 0.14, 0.59 4 0.22 �0.01, 0.45Negative mood 8 0.34 �0.08, 0.76 —Immediate pain 15 0.28 �0.01, 0.57 —Cortisol 7 0.14 �0.10, 0.38 —Blood pressure 5 0.25* 0.03, 0.48 0Heart rate 6 0.41** 0.19, 0.62 0

Multiple-dose effectsTrait anxiety 7 0.75** 0.27, 1.22 0Depression 10 0.62** 0.37, 0.88 0Delayed pain 5 0.31** 0.10, 0.52 3 0.26** 0.07, 0.44

Note. A positive g indicates a reduction for any outcome variable. Dashes indicate data not calculated because of nonsignificance of effect size. CI �confidence interval; L0 � estimate of missing studies resulting from trim and fill procedure.* p � .05. ** p � .01.

13MASSAGE THERAPY META-ANALYSIS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

pain that were lower, on average, than 62% of comparisongroup participants. This finding is consistent with the theorythat MT may promote pain reduction by facilitating restorativesleep, but without data on sleep patterns, this possibility is onlyconjecture.

Reductions of trait anxiety and depression following a course oftreatment were MT’s largest effects. The average MT participantexperienced a reduction of trait anxiety that was greater than 77%of comparison group participants, and a reduction of depressionthat was greater than 73% of comparison group participants. Theseeffects are similar in magnitude to those found in meta-analysesexamining the absolute efficacy of psychotherapy, a more tradi-tional treatment for either condition, in which it is estimated thatthe average psychotherapy client fares better than 79% of un-treated clients (Wampold, 2001, p. 70). Considered together, theseresults indicate that MT may have an effect similar to that ofpsychotherapy.

Moderators

All six moderators that were examined were nonsignificant. Inmost cases, this was not surprising, given that we did not expecteffects to vary according to recipient characteristics and made nopredictions concerning therapist training or laboratory effect.However, it was unexpected that neither the minutes of MT ad-ministered per session nor type of comparison treatment moder-ated effects in a way that was statistically significant.

Minutes of MT administered per session was the only moderatorthat approached the predetermined alpha for statistical signifi-cance. This, combined with the logic that if MT has an effect,longer doses should likely be more potent, leads us to suspect thatour analysis failed to find a relationship because of insufficientstatistical power rather than the true absence of any moderatingeffect. Nevertheless, it must be concluded that this moderator maynot be as important as we predicted, and that even short sessionsof MT can be effective. Future studies could more powerfullyexamine the role of session length by including two levels of thisvariable, something that does not appear to have been done in anystudy to date.

Whether studies used a wait-list equivalent or active/placebocomparison group was not significant for overall effects. Thisfinding does not support the prediction that studies using wait-listequivalent comparison treatments would yield larger effects. Be-cause stronger inferences can be made from within-study compar-isons, we decided to compare this result with those from studiesthat included both an active/placebo and a wait-list equivalentcomparison group within the design. Three studies fitting thiscriterion examined state anxiety as an outcome. Richards (1993),in a study that involved 69 participants, found that wait-list par-ticipants improved significantly less than those who received acombination of muscle relaxation, mental imagery, and relaxingmusic. By contrast, Fraser and Kerr (1993), in a study that in-volved 21 participants, found no statistically significant differencein outcome between two comparison groups, one of which re-ceived attention in the form of conversation (active/placebo), theother of which received no intervention (wait-list equivalent).Similarly, Mueller Hinze (1988), in a study with 48 participants,found no differences in outcome for three comparison groups

including therapeutic touch (active/placebo), transcutaneous elec-trical stimulation without current (active/placebo), and a no-treatment control (wait-list equivalent). As a group, these contrast-ing results seem to agree with the nonsignificant finding in themeta-analysis in suggesting that whether MT is compared with anactive/placebo or wait-list equivalent treatment does not substan-tially influence effects. However, no primary studies that exam-ined MT’s largest effects—on depression and trait anxiety—usedsuch a design; the influence of such a moderator may be moreevident in relation to these more robust effects, and could beexamined in future studies by using both types of comparisongroups.

The prediction that effects would not vary according to the ageor gender of participants was supported. Neither of these recipientcharacteristics was significantly associated with overall effects.Therapist training did not have a significant effect on outcome.This finding, however, should not be used to conclude that trainingis of no consequence. In the present meta-analysis, this variablecould only be dummy coded according to whether a study involveda trained massage therapist, or a layperson trained by a massagetherapist for the purposes of conducting the study. It was notpossible to differentiate the levels of experience various massagetherapists may have had, nor was it possible to know how muchtraining laypersons involved in the studies had received. The onlyconclusion that can be definitively reached from this result is thatlaypersons provided with some training can provide beneficialMT, information that may be valuable to researchers working withlimited resources. No evidence of a laboratory effect was found.

MT Theories

Mixed support for existing theories. It is interesting to notethat, among the theories that are commonly offered to explain MTeffects, the most popular theories are the ones least supported bythe present results. The failure to find a significant effect forimmediate assessment of pain contradicts the theory that MTprovides stimuli that interfere with pain consistent with gate con-trol theory. Reductions in blood pressure and heart rate resultingfrom MT do support the theory that MT promotes a parasympa-thetic response, although, if this theory is true, it would also beexpected that a significant reduction in cortisol levels would haveoccurred, which did not. By contrast, the remaining theories arenot inconsistent with the current results. MT’s effects on stateanxiety, trait anxiety, and depression may come about as a result ofMT’s influence on body chemistry, whereas the ability of a courseof MT treatment to provide lasting pain relief may result from themechanical promotion of circulation and breakdown of adhesions,or from improved sleep promoted by the treatment.

MT from a psychotherapy perspective. Another theory that hasnot previously been put forth may also account for MT effects. MTmay provide benefit in a way that parallels the common-factorsmodel of psychotherapy. Substantial evidence suggests that theconsiderable efficaciousness of psychotherapy results not from anyspecific ingredient of treatment, but rather from the factors that allforms of psychotherapy share (Wampold, 2001). In this model,factors such as a client who has positive expectations for treatment,a therapist who is warm and has positive regard for the client, andthe development of an alliance between the therapist and client are

14 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

considered to be more important than adherence to a specificmodality of psychotherapy. The same model can be extended toMT, given the possibility that benefits arising from it may comeabout more from factors such as the recipient’s attitude towardMT, the therapist’s personal characteristics and expectations, andthe interpersonal contact and communication that take place duringtreatment, as opposed to the specific form of MT used or the siteto which it is applied.

Several of the findings in the present study are consistent withsuch a model applied to MT. The finding that MT has an effect ontrait anxiety and depression that is similar in magnitude to whatwould be expected to result from psychotherapy suggests thepossibility that these different treatments may be more similar thanpreviously considered. Further support comes from the fact thatMT training was not predictive of effects. Possibly, MT effects aremore closely linked with characteristics of the massage providerthat are independent of skill or experience in performing soft tissuemanipulation.

In addition to having similar effects, MT parallels psychother-apy in structure. Both forms of therapy routinely rely on repeated,private interpersonal contact between two persons. Studies con-tributing effects to the trait anxiety and depression outcome cate-gories used treatment protocols similar to those that might bemaintained in short-term psychotherapy, with twice-weekly meet-ings over a span of 5 weeks being most common; other studiesused similar protocols. Interestingly, the length of individual ses-sions in these studies ranged from 15 to 40 min, with 30 min beingthe most common session length. Had these studies used a sessionlength equivalent to the “50-minute hour” that is routine in psy-chotherapy, it is possible that MT’s effect for these variableswould have matched or exceeded that expected of psychotherapy.

Application of such a psychotherapeutic, common-factorsmodel to MT has important ramifications for future research.Different questions need to be asked, different moderators tested,and different comparisons made. Foremost among the questions iswhether MT is as effective as psychotherapy. No study has directlycompared these treatments, a comparison that would be justifiedgiven the finding that some MT effects may be very similar tothose of psychotherapy. Similarly, it could be interesting to deter-mine whether a combination of MT and psychotherapy could besignificantly more effective than either alone. Another criticalissue that needs to be examined is whether these specific MTeffects are enduring. Current studies contributing to these effectsall performed assessments on the final day of treatment, making itimpossible to know if the effects last. Studies that administer acourse of MT treatment should make assessments not only imme-diately after treatment has ended, but also several weeks or monthslater, to determine whether reductions of anxiety, depression, orother conditions are maintained.

Despite the fact that MT is a treatment that relies on interper-sonal contact, no research has attempted to manipulate, or evenmeasure, the kind of psychological interactions that undoubtedlytake place between the provider and recipient of MT. Details worthexamining include (a) the amount and types of communication,both verbal and nonverbal, that take place between massage ther-apist and recipient; (b) the recipient’s and therapist’s expectationsfor whether treatment will be beneficial; (c) the amount of empa-thy perceived by the recipient on behalf of the therapist; (d)

whether the psychological state of the therapist is of importance;and (e) whether personality traits of the therapist, of the recipient,or any interaction between those personality traits influence out-comes. An examination of such personality, process, and thera-peutic relationship variables may reveal that benefiting from MT isjust as much about feeling valued as it is about being kneaded.

Finally, the possibility that MT may provide a significant por-tion of its benefit in a way that parallels psychotherapy has abearing on the selection of comparison treatments used in futureresearch. Viewed from a medical perspective, comparison treat-ments in MT research are thought to function as placebo treat-ments, in that they control for incidental aspects of the treatment(most notably attention in MT research) while withholding what isthought to be the specific effective ingredient (soft tissue manip-ulation). However, the same logic cannot be applied if the treat-ment being examined is thought to be beneficial because of inci-dental aspects, because the double-blind condition favored inmedicine trials, where neither the participants nor the researchersinvolved in the study are aware of who is receiving viable treat-ment and who is receiving the placebo, is logically impossible(Wampold, 2001, p. 129). Those supervising and administeringtreatment in MT research, as in psychotherapy research, are awareof the treatment being delivered and know if it is intended to betherapeutic. This is a critical factor to consider if the treatmentbeing studied relies on the therapist’s beliefs and intentions inorder to be effective. The placebo treatment, derived from medicaltrials intended to examine the effectiveness of specific ingredients,cannot control for the incidental aspects of a treatment such as MT.When a common-factors model is applied to MT, the notion that acomparison treatment such as progressive muscle relaxation con-trols for attention is incorrect. The attention provided to compar-ison group participants is identical in quantity but not in quality,and cannot be expected to function as a control for the attentionreceived by participants in the MT treatment group.

The idea that MT has significant parallels with psychotherapy,and that perspectives gained from psychotherapeutic researchshould be applied to future research, is not meant to suggest thatMT delivers effects entirely by psychological means. Clearly MTis at least partially a physical therapy, and some of its benefitsalmost certainly occur through physiological mechanisms. In fact,one of the most interesting aspects of MT is that it may deliverbenefit in multiple ways; specific ingredients and common factorsmay each play a role, with each being differentially importantdepending on the desired effect. However, whether researcherswish to study MT as a physical therapy, as a psychological one, oras both, new research should examine not merely the effectsresulting from MT, but also the ways in which these effects comeabout. It is only by testing MT theories that a better understandingof this ancient practice will result.

References

References marked with an asterisk indicate studies included in themeta-analysis.

*Abrams, S. M. (1999). Attention-deficit/hyperactivity disordered childrenand adolescents benefit from massage therapy (Doctoral dissertation,University of Miami, 1999). Dissertation Abstracts International, 60,5218.

15MASSAGE THERAPY META-ANALYSIS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

American Massage Therapy Association. (1999a). AMTA definition of mas-sage therapy. Retrieved April 6, 2003, from http://www.amtamassage.org/about/definition.html

American Massage Therapy Association. (1999b). Enhancing your healthwith therapeutic massage. Retrieved April 6, 2003, from http://www.amtamassage.org/publications/enhancing-health.htm

Andersson, S., & Lundeberg, T. (1995). Acupuncture—From empiricismto science: Functional background to acupuncture effects in pain anddisease. Medical Hypotheses, 45, 271–281.

Barbour, L. A., McGuire, D. B., & Kirchhoff, K. T. (1986). Nonanalgesicmethods of pain control used by cancer outpatients. Oncology NursingForum, 13, 56–60.

Barr, J. S., & Taslitz, N. (1970). The influence of back massage onautonomic functions. Physical Therapy, 50, 1679–1691.

Bauer, W. C., & Dracup, K. A. (1987). Physiologic effect of back massagein patients with acute myocardial infarction. Focus on Critical Care, 14,42–46.

Beck, A., Brown, G., Epstein, N., & Steer, R. (1988). An inventory formeasuring clinical anxiety: Psychometric properties. Journal of Consult-ing and Clinical Psychology, 56, 893–897.

Bierenbaum, H., Nichols, M. P., & Schwartz, A. J. (1976). Effects ofvarying session length and frequency in brief emotive psychotherapy.Journal of Consulting and Clinical Psychology, 44, 790–798.

Biggerstaff, B. (2000). S-Plus library DVBID (dvbidlib.exe) [Computersoftware]. Retrieved from http://www.stat.colostate.edu/�bradb/files/

Bollini, P., Pampallona, S., Tibaldi, G., Kupelnick, B., & Munizza, C.(1999). Meta-analysis of dose-effect relationships in randomized clinicaltrials. British Journal of Psychiatry, 174, 297–303.

*Cen, S. Y. (2000) The effect of traditional Chinese therapeutic massageon individuals with neck pain. Unpublished master’s thesis, CaliforniaState University, Northridge.

*Chang, M. Y., Wang, S. Y., & Chen, C. H. (2002). Effects of massage onpain and anxiety during labour: A randomized controlled trial in Taiwan.Journal of Advanced Nursing, 38, 68–73.

Chen, M. L., Lin, L. C., Wu, S. C., & Lin, J. G. (1999). The effectivenessof acupressure in improving the quality of sleep of institutionalizedresidents. Journals of Gerontology: Series A: Medical Sciences, 54,M389–M394.

*Cherkin, D. C., Eisenberg, D., Sherman, K. J., Barlow, W., Kaptchuk,T. J., Street, J., et al. (2001). Randomized trial comparing traditionalChinese medical acupuncture, therapeutic massage, and self-care edu-cation for chronic low back pain. Archives of Internal Medicine, 161,1081–1088.

*Chin, C. C. (1999). The effects of back massage on surgical stressresponses and postoperative pain (Doctoral dissertation, Case WesternReserve University, 1999). Dissertation Abstracts International, 61,776.

Conners, C. K. (1969). A teacher rating scale for use in drug studies withchildren. American Journal of Psychiatry, 126, 884–889.

*Cottingham, J. T., Porges, S. W., & Richmond, K. (1988). Shifts in pelvicinclination angle and parasympathetic tone produced by Rolfing softtissue manipulation. Physical Therapy, 68, 1364–1370.

Curtis, K. A., Roach, K. E., Applegate, E. B., Amar, T., Benbow, C. S.,Genecco, T. D., et al. (1995). Development of the Wheelchair User’sShoulder Pain Index. Paraplegia, 33, 290–293.

*Delaney, J. P., Leong, K. S., Watkins, A., & Brodie, D. (2002). Theshort-term effects of myofascial trigger point massage therapy on car-diac autonomic tone in healthy subjects. Journal of Advanced Nursing,37, 364–371.

Derogatis, L. R. (1983). SCL-90-R administration, scoring and proceduresmanual. Towson, MD: Clinical Psychometric Research.

*Diego, M. A., Field, T., Hernandez-Reif, M., Hart, S., Brucker, B., Field,

T., et al. (2002). Spinal cord patients benefit from massage therapy.International Journal of Neuroscience, 112, 133–142.

*Diego, M. A., Field, T., Hernandez-Reif, M., Shaw, K., Friedman, L., &Ironson, G. (2001). HIV adolescents show improved immune functionfollowing massage therapy. International Journal of Neuroscience, 106,35–45.

Donnelly, C. J., & Wilton, J. (2002). The effect of massage to scars onactive range of motion and skin mobility. British Journal of HandTherapy, 7, 5–11.

Duval, S., & Tweedie, R. (2000). A nonparametric “trim and fill” methodof accounting for publication bias in meta-analysis. Journal of theAmerican Statistical Association, 95, 89–98.

*Dyson-Hudson, T. A., Shiflett, S. C., Kirshblum, S. C., Bowen, J. E., &Druin, E. L. (2001). Acupuncture and Trager psychophysical integrationin the treatment of wheelchair user’s shoulder pain in individuals withspinal cord injury. Archives of Physical Medicine and Rehabilitation, 82,1038–1046.

Eisenberg, D. M., Davis, R. B., Ettner, S. L., Appel, S., Wilkey, S., VanRompay, M., et al. (1998). Trends in alternative medicine use in theUnited States, 1990–1997: Results of a follow-up national survey.Journal of the American Medical Association, 280, 1569–1575.

Elton, D., Stanley, G., & Burrows, G. (1983). Psychological control ofpain. New York: Grune & Stratton.

Ernst, E. (1998). Does post-exercise massage treatment reduce delayedonset muscle soreness? A systematic review. British Journal of SportsMedicine, 32, 212–214.

Fakouri, C., & Jones, P. (1987). Relaxation Rx: Slow stroke back rub.Journal of Gerontological Nursing, 13, 32–35.

Ferrell-Torry, A. T., & Glick, O. J. (1993). The use of therapeutic massageas a nursing intervention to modify anxiety and the perception of cancerpain. Cancer Nursing, 16, 93–101.

Field, T. M. (1998). Massage therapy effects. American Psychologist, 53,1270–1281.

*Field, T., Diego, M., Cullen, C., Hernandez-Reif, M., Sunshine, W., &Douglas, S. (2002). Fibromyalgia pain and substance p decrease andsleep improves after massage therapy. Journal of Clinical Rheumatol-ogy, 8, 72–76.

Field, T., Grizzle, N., Scafidi, F., & Schanberg, S. (1996). Massage andrelaxation therapies’ effects on depressed adolescent mothers. Adoles-cence, 31, 903–911.

Field, T., Hernandez-Reif, M., Hart, S., Theakston, H., Schanberg, S., &Kuhn, C. (1999). Pregnant women benefit from massage therapy. Jour-nal of Psychosomatic Obstetrics and Gynaecology, 20, 31–38.

Field, T. M., Hernandez-Reif, M., Quintino, O., Schanberg, S., & Kuhn, C.(1998). Elder retired volunteers benefit from giving massage therapy toinfants. Journal of Applied Gerontology, 17, 229–239.

*Field, T., Ironson, G., Scafidi, F., & Nawrocki, T., Gonclaves, A., &Burman, I. (1996). Massage therapy reduces anxiety and enhances EEGpattern of alertness and math computations. International Journal ofNeuroscience, 86, 197–206.

Field, T., Morrow, C., Valdeon, C., Larson, S., Kuhn, C., & Schanberg, S.(1992). Massage reduces anxiety in child and adolescent psychiatricpatients. Journal of the American Academy of Child & AdolescentPsychiatry, 31, 124–131.

Field, T., Peck, M., Hernandez-Reif, M., Krugman, S., Burman, I., &Ozment-Schenck, L. (2000). Postburn itching, pain, and psychologicalsymptoms are reduced with massage therapy. Journal of Burn CareRehabilitation, 21, 189–193.

Field, T., Peck, M., Krugman, S., Tuchel, T., Schanberg, S., Kuhn, C., etal. (1998). Burn injuries benefit from massage therapy. Journal of BurnCare and Rehabilitation, 19, 241–244.

Field, T., Quintino, O., Henteleff, T., Wells-Keife, L., & Delvecchio-

16 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Feinberg, G. (1997). Job stress reduction therapies. Alternative Thera-pies, 3, 54–56.

Field, T., Schanberg, S., Kuhn, C., Field, T., Fierro, K., Henteleff, T., et al.(1998). Bulimic adolescents benefit from massage therapy. Adolescence,33, 555–563.

Field, T. M., Sunshine, W., Hernandez-Reif, M., Quintino, O., Schanberg,S., Kuhn, C., et al. (1997). Massage therapy effects on depression andsomatic symptoms in chronic fatigue syndrome. Journal of ChronicFatigue Syndrome, 3, 43–51.

*Fischer, R. L., Bianculli, K. W., Sehdev, H., & Hediger, M. L. (2000).Does light pressure effleurage reduce pain and anxiety associated withgenetic amniocentesis? A randomized clinical trial. Journal of Maternal-Fetal Medicine, 9, 294–297.

*Fraser, J., & Kerr, J. R. (1993). Psychophysiological effects of backmassage on elderly institutionalized patients. Journal of Advanced Nurs-ing, 18, 238–245.

Fritz, S. (2000). Mosby’s fundamentals of therapeutic massage. St. Louis,MO: Mosby.

Glaser, D. (1990). The effects of a massage program on reducing theanxiety of college students (Master’s thesis, San Jose State University,1990). Masters Abstracts International, 29, 263.

*Groer, M., Mozingo, J., Droppleman, P., Davis, M., Jolly, M. L., Boyn-ton, M., et al. (1994). Measures of salivary secretory immunoglobulin Aand state anxiety after a nursing back rub. Applied Nursing Research, 7,2–6.

*Hemenway, C. B. (1993). The effects of massage on pain in labor(Master’s thesis, University of Florida, 1993). Masters Abstracts Inter-national, 33, 515.

*Hernandez-Reif, M., Dieter, J., Field, T., Swerdlow, B., & Diego, M.(1998). Migraine headaches are reduced by massage therapy. Interna-tional Journal of Neuroscience, 96, 1–11.

*Hernandez-Reif, M., Field, T., Field, T., & Theakston, H. (1998). Mul-tiple sclerosis patients benefit from massage therapy. Journal of Body-work and Movement Therapies, 2, 168–174.

*Hernandez-Reif, M., Field, T., Krasnegor, J., & Theakston, H. (2001).Lower back pain is reduced and range of motion increased after massagetherapy. International Journal of Neuroscience, 106, 131–145.

*Hernandez-Reif, M., Field, T., Krasnegor, J., Theakston, H., Hossain, Z.,& Burman, I. (2000). High blood pressure and associated symptomswere reduced by massage therapy. Journal of Bodywork and MovementTherapies, 4, 31–38.

*Hernandez-Reif, M., Field, T., Largie, S., Cullen, C., Beutler, J., Sanders,C., et al. (2002). Parkinson’s disease symptoms are differentially af-fected by massage therapy vs. progressive muscle relaxation: A pilotstudy. Journal of Bodywork and Movement Therapies, 6, 177–182.

*Hernandez-Reif, M., Martinez, A., Field, T., Quintero, O., Hart, S., &Burman, I. (2000). Premenstrual symptoms are relieved by massagetherapy. Journal of Psychosomatic Obstetrics and Gynecology, 21,9–15.

*Hsieh, C. Y., Phillips, R. B., Adams, A. H., & Pope, M. H. (1992).Functional outcomes of low back pain: Comparison of four treatmentgroups in a randomized controlled trial. Journal of Manipulative andPhysiological Therapeutics, 15, 4–9.

Hudson-Cook, N., Tomes-Nicholson, K., & Breen, A. (1989). A revisedOswestry disability questionnaire. In M. O. Roland & J. R. Jennifer(Eds.), Back pain: New approaches to rehabilitation and education (pp.197–204). New York: Manchester University Press.

Hughes, J. E. (1987). Psychological and social consequences of cancer.Cancer Surveys, 6, 455–475.

Hulme, J., Waterman, H., & Hillier, V. F. (1999). The effect of footmassage on patients’ perception of care following laparoscopic steril-ization as day case patients. Journal of Advanced Nursing, 30, 460–468.

Ingram, R. E., & Siegle, G. J. (2002). Contemporary methodological issues

in the study of depression: Not your father’s Oldsmobile. In I. H. Gotlib& C. L. Hammen (Eds.), Handbook of depression (pp. 86–114). NewYork: Guilford Press.

Ironson, G., Field, T., Scafidi, F., Kumar, M., Patarca, R., Price, A., et al.(1996). Massage therapy is associated with enhancement of the immunesystem’s cytotoxic capacity. International Journal of Neuroscience, 84,205–218.

Kovacs, M. (1992). The Children’s Depression Inventory: A self-rateddepression scale for school-aged youngsters. Unpublished manuscript.

Labyak, S. E., & Metzger, B. L. (1997). The effects of effleurage backrubon the physiological components of relaxation: A meta-analysis. Nurs-ing Research, 46, 59–62.

Leak, A. M., Cooper, J., Dyer, S., Williams, K. A., Turner-Stokes, L., &Frank, A. O. (1994). The Northwick Park Neck Pain Questionnairedevised to measure neck pain and disability. British Journal of Rheu-matology, 33, 469–474.

*Leivadi, S., Hernandez-Reif, M. H., Field, T., O’Rourke, M., D’Arienzo,S., Lewis, D., et al. (1999). Massage therapy and relaxation effects onuniversity dance students. Journal of Dance Medicine and Science, 3,108–112.

*Levin, S. R. (1990). Acute effects of massage on the stress response.Unpublished master’s thesis, University of North Carolina, Greensboro.

Lin, E. H., Katon, W. J., Simon, G. E., Von Korff, M., Bush, T. M., Rutter,C. M., et al. (1997). Achieving guidelines for the treatment of depressionin primary care: Is physician education enough? Medical Care, 35,831–842.

Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. ThousandOaks, CA: Sage.

Luscombe, B. (2002, July 29). Massage goes mainstream. Time, 160,48–50.

Malkin, K. (1994). Use of massage in clinical practice. British Journal ofNursing, 3, 292–294.

McNair, D. M., Lorr, M., & Droppleman, L. F. (1971). Profile of MoodStates. San Diego, CA: Educational and Industrial Testing Service.

Melzack, R. (1975). The McGill Pain Questionnaire: Major properties andscoring methods. Pain, 1, 277–299.

Melzack, R. (1987). The short-form McGill Pain Questionnaire. Pain, 30,191–197.

Melzack, R., & Wall, P. D. (1965, November 19). Pain mechanisms: Anew theory. Science, 150, 971–979.

*Menard, M. B. (1995). The effect of therapeutic massage on post surgicaloutcomes (Doctoral dissertation, University of Virginia, 1995). Disser-tation Abstracts International, 57, 276.

Merskey, H., Albe-Fessard, D. G., Bonica, J. J., Carmen, A., Dubner, R.,Kerr, F. W. L., et al. (1979). IASP sub-committee on taxonomy. Pain, 6,249–252.

Moos, R. H. (1968). The development of a menstrual distress question-naire. Psychosomatic Medicine, 30, 853–867.

*Mueller Hinze, M. L. (1988). The effects of therapeutic touch andacupressure on experimentally-induced pain (Doctoral dissertation, Uni-versity of Texas at Austin, 1988). Dissertation Abstracts International,49, 4755.

*Okvat, H. A., Oz, M. C., Ting, W., & Namerow, P. B. (2002). Massagetherapy for patients undergoing cardiac catheterization. AlternativeTherapies in Health and Medicine, 8, 68–75.

Ottenbacher, K. J., Muller, L., Brandt, D., Heintzelman, A., Hojem, P., &Sharpe, P. (1987). The effectiveness of tactile stimulation as a form ofearly intervention: A quantitative evaluation. Developmental and Behav-ioral Pediatrics, 8, 68–76.

Oumeish, O. Y. (1998). The philosophical, cultural, and historical aspectsof complementary, alternative, unconventional, and integrative medicinein the Old World. Archives of Dermatology, 134, 1373–1386.

Phillips, B. N., Martin, R. P., & Meyers, J. (1972). Interventions in relation

17MASSAGE THERAPY META-ANALYSIS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

to anxiety in school. In C. D. Spielberger (Ed.), Anxiety: Vol. 2. Currenttrends in theory and research (pp. 409–464). New York: AcademicPress.

Pinquart, M., & Soerensen, S. (2001). How effective are psychotherapeuticand other psychosocial interventions with older adults? A meta-analysis.Journal of Mental Health and Aging, 7, 207–243.

Popkin, M. K., Callies, A. L., Lentz, R. D., Cohen, E. A., & Sutherland,D. E. (1988). Prevalence of major depression, simple phobia, and otherpsychiatric disorders in patients with long-standing Type-1 diabetesmellitus. Archives of General Psychiatry, 45, 64–68.

*Preyde, M. (2000). Effectiveness of massage therapy for subacute low-back pain: A randomized controlled trial. Canadian Medical AssociationJournal, 162, 1815–1820.

Radloff, L. (1977). The CES–D Scale: A self-report depression scale forresearch in the general population. Applied Psychological Measurement,1, 385–401.

Reed, B. V., & Held, J. (1988). Effects of sequential connective tissuemassage on autonomic nervous system of middle-aged and elderlyadults. Physical Therapy, 68, 1231–1234.

*Rexilius, S. J., Mundt, C. A., Megel, M. E., & Agrawal, S. (2002).Therapeutic effects of massage therapy and healing touch on caregiversof patients undergoing autologous hematopoietic stem cell transplant.Oncology Nursing Forum, 29, E35–E44.

Reynolds, C. R., & Richmond, B. O. (1985). Revised Children’s ManifestAnxiety Scale (RCMAS) manual. Los Angeles: Western PsychologicalServices.

*Richards, K. C. (1993). The effect of a muscle relaxation, imagery, andrelaxing music intervention and a back massage on the sleep andpsychophysiological arousal of elderly males hospitalized in the criticalcare environment (Doctoral dissertation, University of Texas at Austin,1993). Dissertation Abstracts International, 54, 2443.

Rosenthal, R. (1998). Writing meta-analytic reviews. In A. E. Kazdin(Ed.), Methodological issues and strategies in clinical research (2nd ed.,pp. 767–790). Washington, DC: American Psychological Association.

Salvo, S. G. (1999). Massage therapy: Principles and practice. Philadel-phia: Saunders.

Sarafino, E. P. (2002). Health psychology: Biopsychosocial interactions.New York: Wiley.

Scafidi, F., Field, T., Schanberg, S., Bauer, C., Vega-Lahr, N., Garcia, R.,et al. (1986). Effects of tactile/kinesthetic stimulation on the clinicalcourse and sleep/wake behavior of preterm neonates. Infant Behaviorand Development, 13, 91–105.

Schachner, L., Field, T., Hernandez-Reif, M., Duarte, A. M., & Krasnegor,J. (1998). Atopic dermatitis symptoms decreased in children followingmassage therapy. Pediatric Dermatology, 15, 390–395.

*Scherder, E., Bouma, A., & Steen, L. (1998). Effects of peripheral tactilenerve stimulation on affective behavior of patients with probable Alz-heimer’s disease. American Journal of Alzheimer’s Disease, 13, 61–69.

*Shulman, K. R., & Jones, G. E. (1996). The effectiveness of massagetherapy intervention on reducing anxiety in the workplace. Journal ofApplied Behavioral Science, 32, 160–173.

Spielberger, C. D. (1972). Conceptual and methodological issues in anxietyresearch. In C. D. Spielberger (Ed.), Anxiety: Vol. 2. Current trends intheory and research (pp. 481–493). New York: Academic Press.

Spielberger, C. D. (1983). Manual for the State–Trait Anxiety Inventory.Palo Alto, CA: Consulting Psychologists Press.

*Stratford, P. W., Levy, D. R., Gauldie, S., Miseferi, D., & Levy, K.(1989). The evaluation of phonophoresis and friction massage as treat-ments for extensor carpi radialis tendinitis: A randomized controlledtrial. Physiotherapy Canada, 41, 93–99.

Sunshine, W., Field, T. M., Quintino, O., Fierro, K., Kuhn, C., Burman, I.,et al. (1996). Fibromyalgia benefits from massage therapy and transcu-taneous electrical stimulation. Journal of Clinical Rheumatology, 2,18–22.

Tiemens, B. G., Ormel, J., Jenner, J. A., van der Meer, K., Van Os, T. W.,van den Brink, R. H., et al. (1999). Training primary-care physicians torecognize, diagnose and manage depression: Does it improve patientoutcomes? Psychological Medicine, 29, 833–845.

Tiidus, P. M. (1999). Massage and ultrasound as therapeutic modalities inexercise-induced muscle damage. Canadian Journal of Applied Physi-ology, 24, 267–278.

Turner, P. R., Valtierra, M., Talken, T. R., Miller, V. I., & DeAnda, J. R.(1996). Effect of session length on treatment outcome for college stu-dents in brief therapy. Journal of Counseling Psychology, 43, 228–232.

Wampold, B. E. (2001). The great psychotherapy debate. Mahwah, NJ:Erlbaum.

Watson, D. (2000). Mood and temperament. New York: Guilford Press.*Weinrich, S. P., & Weinrich, M. C. (1990). The effect of massage on pain

in cancer patients. Applied Nursing Research, 3, 140–145.Weisz, J. R., Weiss, B., Alicke, M. D., & Klotz, M. L. (1987). Effective-

ness of psychotherapy with children and adolescents: A meta-analysisfor clinicians. Journal of Consulting and Clinical Psychology, 55, 542–549.

*Wendler, M. C. (1999). An investigation of selected outcomes of Tell-ington touch in healthy soldiers (Doctoral dissertation, University ofColorado, 1999). Dissertation Abstracts International, 60, 5439.

White House Commission on Complementary and Alternative MedicinePolicy. (2002, March). White House Commission on Complementaryand Alternative Medicine Policy: Final report. Retrieved May 4, 2003,from http://govinfo.library.unt.edu/whccamp/finalreport.html

*Wilkie, D. J., Kampbell, J., Cutshall, S., Halabisky, H., Harmon, H.,Johnson, L. P., et al. (2000). Effects of massage on pain intensity,analgesics and quality of life in patients with cancer pain: A pilot studyof a randomized clinical trial conducted within hospice care delivery.Hospice Journal, 15, 31–53.

Yyldyz, A., & Sachs, G. S. (2001). Administration of antidepressants:Single versus split dosing: A meta-analysis. Journal of Affective Disor-ders, 66, 199–206.

Received August 8, 2002Revision received June 3, 2003

Accepted June 4, 2003 �

18 MOYER, ROUNDS, AND HANNUM

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.